Add sse4.2 optimized function ws_mempbrk_sse42()

In text protocols, like SIP, lot of time is spend guint8_pbrk(),
assume that text is not binary (no NULs), and use SSE4.2 pcmpistri
instruction.

Also move & rename guint8_pbrk() from tvbuff.c as _ws_mempbrk.

HAVE_SSE42 must be defined to use _ws_mempbrk_sse42() only activaded for
Windows currently.

Change-Id: Ic853d84805bdb6492c4f45d2bcc79a973fd9804e
Reviewed-on: https://code.wireshark.org/review/1730
Reviewed-by: Anders Broman <a.broman58@gmail.com>
This commit is contained in:
Jakub Zawadzki 2014-05-22 23:04:40 +02:00 committed by Anders Broman
parent 6669566199
commit fcb710baec
7 changed files with 292 additions and 17 deletions

View File

@ -281,3 +281,6 @@
#if !defined(QT_VERSION) || !defined(_SSIZE_T_DEFINED)
typedef int ssize_t;
#endif
/* to use define _ws_mempbrk_sse42 if available (checked with cpuinfo) */
#define HAVE_SSE42 1

View File

@ -739,26 +739,17 @@ fast_ensure_contiguous(tvbuff_t *tvb, const gint offset, const guint length)
return NULL;
}
extern const guint8 *ws_mempbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles);
static inline const guint8*
guint8_pbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles, guchar *found_needle)
{
gchar tmp[256] = { 0 };
const guint8 *haystack_end;
const guint8 *result = ws_mempbrk(haystack, haystacklen, needles);
while (*needles)
tmp[*needles++] = 1;
if (result && found_needle)
*found_needle = *result;
haystack_end = haystack + haystacklen;
while (haystack < haystack_end) {
if (tmp[*haystack]) {
if (found_needle)
*found_needle = *haystack;
return haystack;
}
haystack++;
}
return NULL;
return result;
}
@ -2963,6 +2954,12 @@ tvb_get_nstringz0(tvbuff_t *tvb, const gint offset, const guint bufsize, guint8*
gint
tvb_find_line_end(tvbuff_t *tvb, const gint offset, int len, gint *next_offset, const gboolean desegment)
{
#ifdef WIN32
static const char __declspec(align(16)) crlf[] = "\r\n" ;
#else
static const char crlf[] __attribute__((aligned(16))) = "\r\n" ;
#endif
gint eob_offset;
gint eol_offset;
int linelen;
@ -2981,7 +2978,7 @@ tvb_find_line_end(tvbuff_t *tvb, const gint offset, int len, gint *next_offset,
/*
* Look either for a CR or an LF.
*/
eol_offset = tvb_pbrk_guint8(tvb, offset, len, "\r\n", &found_needle);
eol_offset = tvb_pbrk_guint8(tvb, offset, len, crlf, &found_needle);
if (eol_offset == -1) {
/*
* No CR or LF - line is presumably continued in next packet.

View File

@ -52,6 +52,10 @@ if NEED_STRPTIME_LO
wsutil_optional_objects += @STRPTIME_LO@
endif
## if SSE42_SUPPORTED
wsutil_optional_objects += libwsutil_sse42.la
## endif
include ../Makefile.am.inc
include Makefile.common
@ -62,6 +66,8 @@ if HAVE_WARNINGS_AS_ERRORS
AM_CFLAGS += -Werror
endif
noinst_LTLIBRARIES = libwsutil_sse42.la
lib_LTLIBRARIES = libwsutil.la
# http://www.gnu.org/software/libtool/manual/html_node/Updating-version-info.html
libwsutil_la_LDFLAGS = -version-info 0:0:0 @LDFLAGS_SHAREDLIB@
@ -74,6 +80,11 @@ libwsutil_la_SOURCES = \
$(LIBWSUTIL_SRC) \
$(LIBWSUTIL_INCLUDES)
libwsutil_sse42_la_SOURCES = \
ws_mempbrk_sse42.c
libwsutil_sse42_la_CFLAGS = $(AM_CFLAGS) -msse4.2
EXTRA_libwsutil_la_SOURCES= \
inet_aton.c \
inet_aton.h \

View File

@ -60,6 +60,7 @@ LIBWSUTIL_SRC = \
tempfile.c \
time_util.c \
type_util.c \
ws_mempbrk.c \
u3.c \
unicode-utils.c

View File

@ -27,7 +27,8 @@ OBJECTS = file_util.obj \
inet_pton.obj \
$(LIBWSUTIL_SRC:.c=.obj) \
strptime.obj \
wsgetopt.obj
wsgetopt.obj \
ws_mempbrk_sse42.obj
# For use when making libwsutil.dll
libwsutil.lib: libwsutil.dll

74
wsutil/ws_mempbrk.c Normal file
View File

@ -0,0 +1,74 @@
/* ws_mempbrk.c
*
* Wireshark - Network traffic analyzer
* By Gerald Combs <gerald@wireshark.org>
* Copyright 1998 Gerald Combs
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include "config.h"
#include <glib.h>
#include "ws_symbol_export.h"
#include "ws_cpuid.h"
#ifdef HAVE_SSE42
extern const char *_ws_mempbrk_sse42(const char* haystack, size_t haystacklen, const char *needles);
#endif
const guint8 *_ws_mempbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles);
const guint8 *
_ws_mempbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles)
{
gchar tmp[256] = { 0 };
const guint8 *haystack_end;
while (*needles)
tmp[*needles++] = 1;
haystack_end = haystack + haystacklen;
while (haystack < haystack_end) {
if (tmp[*haystack])
return haystack;
haystack++;
}
return NULL;
}
WS_DLL_PUBLIC const guint8 *
ws_mempbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles)
{
#ifdef HAVE_SSE42
guint32 CPUInfo[4];
guint32 bSSE42Extensions;
/*const int has_sse42 = 1;*/ /* XXX, use CPUID */
#endif
if (*needles == 0)
return NULL;
#ifdef HAVE_SSE42
ws_cpuid(CPUInfo, 1);
bSSE42Extensions = (CPUInfo[2] & 0x100000);
if (haystacklen >= 16 && bSSE42Extensions)
return _ws_mempbrk_sse42(haystack, haystacklen, needles);
#endif
return _ws_mempbrk(haystack, haystacklen, needles);
}

188
wsutil/ws_mempbrk_sse42.c Normal file
View File

@ -0,0 +1,188 @@
/* strcspn with SSE4.2 intrinsics
Copyright (C) 2009-2014 Free Software Foundation, Inc.
Contributed by Intel Corporation.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#include <glib.h>
#ifdef WIN32
#include <tmmintrin.h>
#include <stdint.h>
#endif
#include <nmmintrin.h>
#include <string.h>
extern const guint8 *_ws_mempbrk(const guint8* haystack, size_t haystacklen, const guint8 *needles);
const char *_ws_mempbrk_sse42(const char* haystack, size_t haystacklen, const char *needles);
/* Helper for variable shifts of SSE registers.
Copyright (C) 2010 Free Software Foundation, Inc.
*/
static const int8_t ___m128i_shift_right[31] =
{
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
};
static inline __m128i
__m128i_shift_right (__m128i value, unsigned long int offset)
{
/* _mm_loadu_si128() works with unaligned data, cast safe */
return _mm_shuffle_epi8 (value,
_mm_loadu_si128 ((__m128i *) (void *) (___m128i_shift_right + offset)));
}
/* We use 0x2:
_SIDD_SBYTE_OPS
| _SIDD_CMP_EQUAL_ANY
| _SIDD_POSITIVE_POLARITY
| _SIDD_LEAST_SIGNIFICANT
on pcmpistri to compare xmm/mem128
0 1 2 3 4 5 6 7 8 9 A B C D E F
X X X X X X X X X X X X X X X X
against xmm
0 1 2 3 4 5 6 7 8 9 A B C D E F
A A A A A A A A A A A A A A A A
to find out if the first 16byte data element has any byte A and
the offset of the first byte. There are 3 cases:
1. The first 16byte data element has the byte A at the offset X.
2. The first 16byte data element has EOS and doesn't have the byte A.
3. The first 16byte data element is valid and doesn't have the byte A.
Here is the table of ECX, CFlag, ZFlag and SFlag for 2 cases:
1 X 1 0/1 0
2 16 0 1 0
3 16 0 0 0
We exit from the loop for cases 1 and 2 with jbe which branches
when either CFlag or ZFlag is 1. If CFlag == 1, ECX has the offset
X for case 1. */
const char *
_ws_mempbrk_sse42(const char *s, size_t slen, const char *a)
{
const char *aligned;
__m128i mask;
int offset;
offset = (int) ((size_t) a & 15);
aligned = (const char *) ((size_t) a & -16L);
if (offset != 0)
{
int length;
/* Load masks. */
/* cast safe - _mm_load_si128() it's 16B aligned */
mask = __m128i_shift_right(_mm_load_si128 ((__m128i *) (void *) aligned), offset);
/* Find where the NULL terminator is. */
length = _mm_cmpistri (mask, mask, 0x3a);
if (length == 16 - offset)
{
/* There is no NULL terminator. */
__m128i mask1 = _mm_load_si128 ((__m128i *) (void *) (aligned + 16));
int index = _mm_cmpistri (mask1, mask1, 0x3a);
length += index;
/* Don't use SSE4.2 if the length of A > 16. */
if (length > 16)
return _ws_mempbrk(s, slen, a);
if (index != 0)
{
/* Combine mask0 and mask1. We could play games with
palignr, but frankly this data should be in L1 now
so do the merge via an unaligned load. */
mask = _mm_loadu_si128 ((__m128i *) (void *) a);
}
}
}
else
{
int length;
/* A is aligned. (cast safe) */
mask = _mm_load_si128 ((__m128i *) (void *) a);
/* Find where the NULL terminator is. */
length = _mm_cmpistri (mask, mask, 0x3a);
if (length == 16)
{
/* There is no NULL terminator. Don't use SSE4.2 if the length
of A > 16. */
if (a[16] != 0)
return _ws_mempbrk(s, slen, a);
}
}
offset = (int) ((size_t) s & 15);
aligned = (const char *) ((size_t) s & -16L);
if (offset != 0)
{
/* Check partial string. cast safe it's 16B aligned */
__m128i value = __m128i_shift_right (_mm_load_si128 ((__m128i *) (void *) aligned), offset);
int length = _mm_cmpistri (mask, value, 0x2);
/* No need to check ZFlag since ZFlag is always 1. */
int cflag = _mm_cmpistrc (mask, value, 0x2);
int index = _mm_cmpistri (value, value, 0x3a);
if (cflag)
return s + length;
/* Find where the NULL terminator is. */
if (index < 16 - offset)
{
/* fond NUL @ 'index', need to switch to slower mempbrk */
return _ws_mempbrk(s + index + 1, slen - index - 1, a); /* slen is bigger than 16 & index < 16 so no undeflow here */
}
aligned += 16;
slen -= (16 - offset);
}
else
aligned = s;
while (slen >= 16)
{
__m128i value = _mm_load_si128 ((__m128i *) (void *) aligned);
int index = _mm_cmpistri (mask, value, 0x2);
int cflag = _mm_cmpistrc (mask, value, 0x2);
int zflag = _mm_cmpistrz (mask, value, 0x2);
if (cflag)
return aligned + index;
if (zflag)
{
/* found NUL, need to switch to slower mempbrk */
return _ws_mempbrk(aligned, slen, a);
}
aligned += 16;
slen -= 16;
}
/* XXX, use mempbrk_slow here? */
return _ws_mempbrk(aligned, slen, a);
}